Cohort Analysis in Epidemiology

Epidemiologic cohort studies typically involve the follow-up of large population groups over many years to ascertain the effects of environmental exposures on the outbreak of illness and the age and cause of death. An efficient method of analysis is to fit Poisson regression models to grouped data consisting of a multidimensional classification of disease cases and person-years of observation by discrete categories of age, calendar period, and various aspects of exposure. Extension of these models for use with disease rates and exposure variables that vary continuously with age or time leads to the well-known proportional hazards model. Incorporation of external standard rates is more likely to improve the estimates of exposure effects in additive or excess risk models than in multiplicative or relative risk situations. Examples are provided of the maximum likelihood fitting of such models to data from cohort studies of British doctors and Montana smelter workers. The discussion considers the choice between models and certain problems that may arise when attempting to fit nonmultiplicative relationships.